JP2019537394A - Site detection - Google Patents

Abstract

The examples of the present disclosure describe systems and methods for on-site sensing. In aspects, a mobile device that includes a set of sensors can collect and store sensor data in response to detected movement events or user interaction data. The sensor data can be used to generate a set of candidate sites corresponding to the location of the mobile device. Candidate sites can be provided to a site detection system. The site detection system can process the candidate sites to generate a set of features. The set of features can be applied to one or more probability models and / or used to generate those probability models. The probabilistic model can generate a reliability metric for each of the candidate sites. In some embodiments, the top "N" sites can be selected from the set. The top "N" locations can then be presented to a user and / or used to effect one or more actions.

Description

CROSS REFERENCE TO RELATED APPLICATIONSThis application was filed as a PCT international patent application on September 14, 2017, and was filed on August 29, 2017, entitled VENUE DETECTION. Claims priority to application No. 15 / 689,683, which is filed in US Provisional Application No. / 395,827, the disclosures of which are incorporated herein by reference in their entirety.

  The on-site detection system enables identification related to mobile device visit patterns. In many cases, field sensing systems rely almost exclusively on periodic geographic coordinate system data (eg, latitude, longitude, and / or elevation coordinates) to determine the location of a mobile device. For example, geographic coordinate system data can be used to determine whether a detected outage by a mobile device is correlated with a check-in event by a mobile device user at a particular location. However, almost exclusively use of geographic coordinate system data can result in inaccurate field detection.

  It is with respect to these and other general considerations that the embodiments disclosed herein have been made. Also, although relatively specific issues may be discussed, the example should not be limited to solving specific problems identified in the background or elsewhere in this disclosure. I want to be understood.

  The examples of the present disclosure describe systems and methods for on-site sensing. In aspects, a mobile device that includes a set of sensors may collect and store sensor data from the set of sensors in response to detecting a movement event or user interaction data. The sensor data can be used to generate a set of candidate sites corresponding to the location of the mobile device. Candidate sites can be provided to a site detection system. The site detection system can process the candidate sites to generate a set of features. The set of features may be applied to one or more probability models and / or used to generate those probability models. The probabilistic model can generate a reliability metric for each of the candidate sites. In some embodiments, the top "N" sites can be selected from the set. The top "N" sites can then be presented to a user and / or used to effect one or more actions.

  This "Summary of the Invention" is provided to introduce selected ones of the concepts in a simplified form, which concepts are described below in the "Detailed Description of the Invention". It is further described. This Summary is not intended to identify key or essential features of the claimed subject matter and is used to limit the scope of the claimed subject matter. Nor is it intended to be. Additional aspects, features, and / or advantages of the examples will be set forth in part in the description that follows, and in part will be obvious from the description, or may be learned by practice of the present disclosure. is there.

  Non-limiting and non-exhaustive examples are described with reference to the following figures.

FIG. 1 illustrates an overview of an exemplary system for in-situ sensing described herein. FIG. 3 illustrates an example input processing unit for performing the on-site sensing described herein. FIG. 4 illustrates an exemplary method for performing on-site sensing as described herein. FIG. 3 illustrates an example of a suitable operating environment in which one or more of these embodiments may be implemented.

  Various aspects of the disclosure are more fully described hereinafter with reference to the accompanying drawings, which form a part hereof, and in which are shown certain illustrative aspects. ing. However, the various aspects of the disclosure may be embodied in many different forms and should not be construed as limited to the aspects set forth herein; The embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of these embodiments to those skilled in the art. Aspects can be implemented as a method, system, or device. Thus, aspects may take the form of a hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Therefore, the following detailed description should not be taken in a limiting sense.

  This disclosure describes systems and methods for on-site sensing. In aspects, site detection may be described in connection with passive visit detection. Passive visit detection refers to the use of implicit indicia to identify whether or not a mobile device (or its user) is visiting or has visited a particular site or geographic location. Passive visit detection, in contrast to active visit detection using explicit user signals (such as on-site check-in data confirmed by the user), is a passive visit detection of data from various sensors on the mobile device. It depends on a collective. Analysis of data from various mobile device sensors increases the accuracy and efficiency of visit detection over traditional approaches that only use geographic coordinate system data. Although the on-site detection techniques described herein are described in the context of passive visit detection, it is noted that such techniques may be applied in other contexts. The trader will recognize.

  In aspects, the mobile device can include one or more sensors. Exemplary sensors can include GPS sensors, Wi-Fi sensors, proximity sensors, accelerometers, ambient temperature sensors, gyroscopes, optical sensors, magnetometers, hall sensors, acoustic sensors, touch screen sensors, and the like. . The mobile device can monitor data sensed by the sensor as the mobile device is being used and / or carried by the user. The mobile device can also detect events such as a mobile event, a purchase event, an information distribution event, a field check-in event, and the like. In some aspects, detection of the event can cause sensor data to be collected and processed. Processing the sensor data can include analyzing the sensor data to identify one or more sites and / or associated site data. The identified site / site data can be analyzed to identify a set of candidate sites corresponding to events and / or user activities in the geofenced area. "Setting a geofence" as used herein can refer to dynamically creating a virtual boundary around a geographic location of a mobile device. The set of candidate sites can be analyzed to identify a set of features. The set of features can be used to generate one or more feature vectors for each candidate site. A feature vector, as used herein, can refer to an n-dimensional vector of numerical features representing one or more objects. The feature vector is applied to or used to generate one or more data evaluation utilities, such as decision logic, one or more rule sets, or machine learning models Is possible. A model, as used herein, is used to identify a probability distribution over one or more character sequences, classes, objects, result sets, or events, and / or one or more predictors. Can refer to a predictive or statistical language model that can be used to predict response values from In some examples, the model can be a rule-based model, a machine learning regressor, a machine learning classifier, a neural network, and so on.

  In aspects, the data evaluation utility may use the feature vector to generate a reliability metric for the list of candidate sites. The reliability metric may indicate the likelihood that the candidate site corresponds to a particular set of coordinates or event. In some examples, the list of candidate sites can be ranked according to one or more criteria (such as reliability metrics) and presented via a user interface. In some aspects, the list of candidate sites can be calibrated to increase the accuracy and / or reliability of the candidate site selection or to reduce the number of candidate site selections. Calibration may include the use of heuristics and one or more threshold comparisons. In some examples, the calibration may result in the selection of the top "N" sites from a list of candidate sites. One or more of the top "N" sites can be presented to the user via the display interface. The display interface can provide for receiving one or more user selections. User selection can be used to further calibrate the list of top "N" sites.

  Thus, the present disclosure enables, among other examples, to detect passive visit events, enable site-based geofencing, and collect site / location data using various mobile device sensors Enable the collection of site and user data from one or more data stores, generate site-based feature vectors, apply machine learning techniques to identify probabilities for lists of sites Using site-based negative information to refine site selection; improving the measurement of relative visits between sites; calibrating the site list to remove low-probability sites Assessing the accuracy of stochastic models using objective functions, and applications / services utilizing the examples of this disclosure Including improved efficiency and quality for but are not limited to: providing a plurality of technical benefits.

  FIG. 1 shows an overview of an exemplary system for in-situ sensing as described herein. The exemplary system 100 presented is a combination of interdependent components that interact to form an integrated whole for a field sensing system. The components of the system can be hardware components or software implemented on and / or executed by the hardware components of the system. In some examples, system 100 includes hardware components (e.g., used to run / run an operating system (OS)) and software components running on hardware (e.g., applications, application programming). Interfaces (APIs), modules, virtual machines, runtime libraries, etc.). In one example, the example system 100 can provide an environment for software components to operate, follow a set of constraints for operation, and utilize the resources or facilities of the system 100, In some cases, a component may be software (eg, an application, a program, a module, etc.) running on one or more processing devices. For example, software (e.g., applications, operating instructions, modules, etc.) may be a computer, mobile device (e.g., smartphone / phone, tablet, laptop, personal digital assistant (PDA), etc.), and / or any other electronic device And so on. For an example of a processing device operating environment, see the exemplary operating environment shown in FIG. In other examples, components of the systems disclosed herein may be distributed across multiple devices. For example, the input can be entered on a client device and the information can be processed or accessed from other devices in the network, such as one or more server devices.

  As an example, system 100 includes a client device 102, a distributed network 104, a site analysis system 106, and storage 108. One skilled in the art will appreciate that the scale of a system, such as system 100, can vary and can include more or fewer components than described in FIG. In some examples, interfacing between components of system 100 may be remote, for example, if components of system 100 can be distributed across one or more devices in a distributed network. Can occur.

  Client device 102 can be configured to collect sensor data associated with one or more sites. In aspects, the client device 102 may include or have access to one or more sensors. These sensors provide sensor data about the client device 102, such as GPS coordinates and geolocation data, location data (such as horizontal and / or vertical accuracy), Wi-Fi information, OS information and settings, hardware information, signal strength , Can be operable to detect and / or generate accelerometer data, time information, and the like. Client device 102 is capable of collecting and processing sensor data. In some aspects, the client device 102 may collect and / or process sensor data in response to detecting an event or satisfying one or more criteria. For example, sensor data may be collected in response to a detected outage by client device 102. In some examples, detecting an outage may include one or more machine learning techniques or algorithms, such as an Expectation Maximization (EM) algorithm, a Hidden Markov Model (HMM), a Viterbi algorithm, a forward backward algorithm, It can include the use of a fixed lag smoothing algorithm, a Baum Welch algorithm, and the like. Collecting sensor data may include aggregating data from various sensors, organizing those data by one or more criteria, and / or a data store accessible to client device 102 (shown in FIG. And storing those sensor data. The collected sensor data can be provided (or accessible by) an analysis utility, such as a field analysis system 106, via the distributed network 104.

  The site analysis system 106 can be configured to create one or more site lists. In embodiments, the site analysis system 106 may have access to one or more sets of sensor data. The site analysis system 106 can process the sensor data to identify a candidate set of sites. Processing the sensor data may include analyzing the sensor data to identify one or more locations of the client device 102. One or more locations of the client device 102 can be applied to a classification algorithm, such as a k-nearest neighbor algorithm, to fetch a candidate set of scenes. A k-nearest neighbor algorithm, as used herein, is where an object is classified by majority of its neighbors and the object is placed in the class that is the most common among its k nearest neighbors. It is possible to refer to the classification (or regression) technique that is assigned. In some examples, the candidate set of sites generated using the classification algorithm may include sites that are near or within a certain proximity of the location reported for the client device 102. is there.

  In-situ analysis system 106 can be further configured to generate one or more feature vectors. In aspects, the site analysis system 106 may generate a feature vector for one or more sites in a candidate set of sites. Generating the feature vector can include combining sensor data associated with the scene with information from one or more knowledge sources. In some examples, the information may include previous and / or predicted site and user data. In-situ analysis system 106 may evaluate the feature vectors using one or more probabilistic data models or prediction algorithms, for example, a decision tree. A non-exhaustive list of decision tree learning techniques includes classification trees, regression trees, boosted trees, bootstrap aggregated trees, and rotation forests. In some examples, evaluating the feature vector may include using one or more gradient boosting techniques. Gradient boosting, as used herein, can refer to one or more machine learning techniques for classification and regression, which are in the form of sets of weak predictive models. Create a predictive model. Gradient boosting involves factors such as site age, site popularity, proximity to other sites, historical accuracy of site selection / selection, previous site visits, implicit / explicit user feedback, etc. It is possible to incorporate. Evaluation of the feature vector can lead to the calculation of a score or probability for a site in the candidate set of sites. These scores / probabilities can be assigned to a feature vector (or its features) or to the corresponding sites in the candidate set of sites. Such a score / probability may indicate the likelihood that the scene corresponds to the location data reported by the client device 102. The candidate set of sites, feature vectors and associated feature data, and / or scores / probabilities may be stored in a data store, such as storage 108.

  The site analysis system 106 can be further configured to calibrate a candidate set of sites. In aspects, calibrating the candidate set of sites may include evaluating a score / probability for each candidate site against a threshold. Sites that have a score / probability below the threshold can be removed from the candidate set of sites or reduced in some manner. In some examples, identifying the threshold may include evaluating one or more sets of heuristics against a candidate set of sites to identify a set of site visit distributions. . The site visit distribution may include one or more of the sites in the candidate set of sites and / or one or more sites associated with the reported location of the client device 102. The site analysis system 106 may compare the set of site visit distributions (or aggregated site visit distributions) to a set of labeled and / or unlabeled training data. Based on the comparison, a threshold is identified that identifies sites that are unlikely to correspond to the location data reported by the client device 102. In some aspects, calibrating the candidate set of sites may further include identifying a list of the top "N" sites. The list of top "N" sites may represent the site candidates most likely to correspond to the location of the mobile device. In some examples, identifying a list of the top "N" sites may include evaluating a score / probability for each candidate site against a threshold. For example, the top three sites (e.g., "N" = 3) with the highest score / probability may be selected. In at least one example, identifying the list of the top "N" locations can further include receiving user-selected data. For example, a set of sites can be presented to a user via an interface component accessible to the site analysis system 106. Site analysis system 106 may receive a selection of one or more sites in a set of sites. The selected sites may represent the top "N" sites. Alternatively, the selected site may represent a site that should be removed from the list of top "N" sites, or an adjustment to the ranking of the list of top "N" sites. A list of the top "N" sites can be stored in storage 108 and / or presented via a display interface.

  In-situ analysis system 106 can be further configured to evaluate the accuracy of one or more probabilistic models. In aspects, the site analysis system 106 may have access to a set of training data. Training data can include, for example, previously received input data (eg, sensor data, feature vectors, corresponding feature data, etc.) and / or specific or selected field outputs. The site analysis system 106 can apply the input data to one or more probabilistic models. The output of each probability model can be provided to an objective function that is used to identify the accuracy or validity of one or more probability models. In some examples, the objective function can be defined in a myriad of ways using various parameters. The objective function can calculate a score that represents how close the output of the probabilistic model matches the corresponding output in the training data. Based on the calculated scores, the site analysis system 106 can rank one or more probabilistic models. In one example, the highest ranking probability model can be selected for use when a set of feature vectors is received.

  FIG. 2 shows an overview of an exemplary input processing device 200 for in-situ sensing as described herein. Site detection techniques implemented by the input processing device 200 may include the site detection techniques and content described in FIG. In the alternative, a single system (including one or more components such as a processor and / or memory) may perform the processes described in systems 100 and 200, respectively.

  Referring to FIG. 2, the input processing unit 200 may include a retrieval engine 202, a feature engine 204, a stochastic model engine 206, a user interface 208, and a calibration engine 210. Retrieval engine 202 may be configured to generate one or more feature sets for the set of sensor data. In aspects, the retrieval engine 202 may be able to access a set of sensor data collected from a mobile device. The sensor data can represent input from a user or a physical environment associated with the mobile device. The sensor data can include inputs collected from various sensors of the mobile device (or accessible to the mobile device). In some aspects, the retrieval engine 202 can process the sensor data in response to detecting an event or satisfying one or more criteria. For example, the retrieval engine 202 may receive, or otherwise detect, an indication that the mobile device has stopped moving. Detecting outages involves sampling sensor data over time, characterizing those sensor data, and applying an outage detection model to those characterized data to determine if the mobile device has stopped moving Can be specified. Processing the sensor data can include analyzing the sensor data for one or more individual sensors or sensor types to identify the site and / or associated site data. In aspects, the site data can be used to retrieve a set of candidate sites. For example, the field data can be applied to a classification algorithm, such as a k-nearest neighbor algorithm. The classification algorithm can identify a list of candidate sites that are within a particular proximity or density distribution of the location reported for the mobile device. In another example, a set of geographic coordinates (or other geolocation identifier) can be provided to a site-specific utility, such as a geographic mapping service. In such an example, the number of sites may depend on the size of the display area presenting those sites.

  The feature engine 204 can be configured to generate a feature vector for the list of candidate sites. In aspects, the feature engine 204 may have access to a list of candidate sites. Feature engine 204 may correlate each candidate site with information associated with the candidate site. For example, the candidate sites may be correlated to popular nearby sites, sites with similar names or themes, or other related sites. Site data for the candidate site and the correlated site data can be identified and used to generate a feature vector for each identified site. In one example, generating a feature vector includes organizing similar features or feature categories, normalizing corresponding feature values, and adding the normalized feature values to the feature vector. Can be included. In aspects, the feature vector may indicate how well the sensor data for the site matches previous data for the site. As an example, the mobile device can perform a Wi-Fi scan during a detected outage. The Wi-Fi scan data can be compared to the Wi-Fi scan data of each candidate site to generate a feature score. The feature score can indicate how well the Wi-Fi data matches each of the candidate sites.

  The probabilistic model engine 206 can be configured to generate and / or train one or more models. In aspects, the probabilistic model engine 206 may have access to one or more feature vectors. The probabilistic model engine 206 can apply one or more gradient boosting techniques to generate a probabilistic model that includes a set of decision trees. Gradient boosting techniques can incorporate factors such as age of the site, site popularity, proximity to other sites, historical accuracy of site selection / selection, previous site visits, and the like. In at least one example, gradient boosting techniques can also incorporate explicit user feedback. For example, a result set that includes explicit user confirmation or negation of the site location may be used to generate the probability model 206. In some aspects, the probability model may have access to one or more feature vectors. The feature vector may include feature data and / or a feature score for a candidate set of sites. The probabilistic model can receive a feature vector as input and can output a confidence score or probability for each site in the candidate set of sites. The confidence score / probability can indicate the likelihood that the scene corresponds to a particular set of coordinates or event. In at least one example, the probabilistic model can generate a probability distribution for sites in a candidate set of sites. The probability distribution may assign a probability or reliability score to each site in the site list, for example, using a probability density function.

  In aspects, the stochastic model engine 206 can be further configured to evaluate the accuracy of one or more stochastic models. The probabilistic model engine 206 may include, for example, training including previously received input data (eg, sensor data, feature vectors, feature scores, etc.), explicit user feedback, labeled field outputs, field-specific analysis, etc. It is possible to access a set of data. The probabilistic model engine 206 can apply the input data to one or more probabilistic models. The probabilistic model may have been previously generated, recently generated, or may be generated as part of evaluation by the probabilistic model engine 206. The output of each probability model can be provided to an objective function that is used to identify the accuracy or validity of one or more probability models. In some examples, the objective function can define criteria and / or parameters to be evaluated. The objective function can define the number and type of stochastic models evaluated. In some aspects, the objective function may compare the output of each probability model with the output in the training data. Based on the comparison, the objective function can calculate a score that indicates how close the probabilistic model output matches the training data output. The highest ranking probabilistic model can be recorded and selected for use if the feature vector is subsequently received.

  The user interface 208 can be configured to present a venue list. In aspects, the user interface 208 may allow a user to view, navigate, and manipulate the site listing and / or data associated therewith. In some examples, the user interface 208 may present the site listing as a textual list, a graphical object (eg, a geographic map or similar visualization), a report, or the like. The user interface 208 may also provide access to settings or configurations for interacting with the site list. User interface 208 can be further configured to provide training data to stochastic model engine 206. In aspects, the user interface 208 provides access to one or more sets of training data. The training data can be located locally on the input processing unit 200 or on a remote device. In some examples, the training data may be a labeled or unlabeled input signal. The labeled data may include the sensor values as well as the corresponding labeled events and locations. Training data can be applied to the stochastic model to train the stochastic model to create a more accurate or relevant site list. In some examples, creating a more accurate or relevant site list is to identify the parameters / features that are most relevant or descriptive within the set of labeled data Can be included. Such identification may include applying weights to the parameters / features and / or removing the parameters / features from the site list analysis.

  Calibration engine 210 may be configured to filter the site list. In aspects, the calibration engine 210 may have access to one or more site listings. The calibration engine 210 can apply one or more processing techniques to the site list to increase the accuracy and / or reliability of the site selection or reduce the number of site selections. For example, the calibration engine 210 may compare one or more probabilities or confidence scores associated with a candidate set of sites to a threshold. The threshold value may represent a minimum probability / reliability score for a site candidate that should be included in the site candidate set. As a specific example, sites having a probability / reliability score of less than 1% can be removed from the candidate set of sites. In aspects, the threshold may be specified or selected, for example, by applying a rule set, a set of heuristics, or other analytical techniques to a candidate set of sites to generate a site visit distribution. It is. The site visit distribution can identify the frequency with which one or more sites reported visits during a period of time. In some aspects, the calibration engine 210 can compare the site visit distribution with the set training data. The training data can include labeled or unlabeled site data, such as site, check-in information, user data and signals, probability / reliability scores, and the like. The training data can be used to assess the probability / reliability score and / or visit distribution of the site. Based on the evaluation, a threshold can be selected. The threshold may indicate a certain probability / reliability score, below which the scene is unlikely to correspond to location data reported by the mobile device. The threshold can be used to identify a list of the top "N" sites. The list of top “N” sites may represent the site candidates most likely to correspond to the location of the mobile device. In some examples, the top "N" sites can be identified by selecting a specific number of site results, selecting site results above a certain probability threshold, and so on. .

  Having described various systems that can be employed by aspects disclosed herein, the present disclosure can then be implemented by one or more aspects that can be implemented by various aspects of the present disclosure. Describe multiple methods. In aspects, the method 300 may be performed by an exemplary system, such as the system 100 of FIG. In some examples, method 300 can be performed on a device such as input processing unit 200 that includes at least one processor configured to store and execute operations, programs, or instructions. . However, the method 300 is not limited to such an example. In other examples, the method 300 can be performed on an application or service for performing site detection. In at least one example, method 300 can be performed by one or more components of a distributed network, such as a web service / distributed network service (e.g., a cloud service) (e.g., implemented by a computer). Operations).

  FIG. 3 illustrates an exemplary method 300 for in-situ sensing as described herein. The example method 300 begins at operation 302, where sensor data may be received. In aspects, sensor data from one or more sensors on (or associated with) a mobile device, such as client device 102, can be monitored or collected. Sensor data includes GPS coordinates and geolocation data, location data (eg, horizontal accuracy data, vertical accuracy data, etc.), Wi-Fi data, over-the-air (OTA) data (eg, Bluetooth® data, Wireless communication (NFC) data, etc., OS information and settings, hardware / software information, signal strength data, movement information (e.g., acceleration, time and direction data), etc. . Sensor data can be continuously, intermittently, on demand, or detected stoppages, recognizable changes in travel speed and / or direction, check-ins, purchase events, messages received by mobile devices, etc. It can be collected depending on the fulfillment of one or more criteria.

  In aspects, sensor data can be monitored to identify relevant data points, time frames, travel events, and / or site information. For example, sensor data can be monitored as the mobile device is moving along a storefront. A stop detection utility can monitor sensor data received by the mobile device. During the first time period, the outage detection utility sends a Wi-Fi signal (e.g., `` Store A '' Wi-Fi network) for 10 consecutive polling cycles (e.g., a 10 minute time period). It is possible to analyze sensor data, including detection. The outage detection utility can determine that the mobile device was continuously in proximity to store A because the Wi-Fi signal was detected during each of the ten polling cycles. Therefore, the stop detection utility can specify the visit state of “visiting” for store A. In response to the visiting state of “visiting,” the outage detection utility may collect and store sensor data during the first time period. During the second time period, the outage detection utility provides a sensor data including multiple Wi-Fi signals (e.g., detection of the "Store A" and "Store B" Wi-Fi networks) and corresponding signal strengths. Can be analyzed. Based on the sensor data, the stop detection utility can specify that the mobile device was close to store B but the mobile device did not actually enter the store. For example, the Wi-Fi network device for Store B may be 55 feet inside the storefront door. The device measures the signal strength of the `` Store B '' Wi-Fi network at -80 dBm at a radius of 55 feet from the Wi-Fi network device, -70 dBm at a radius of 25 feet from the Wi-Fi network device, and Wi-Fi. At a radius of 5 feet from a Fi network device, it can be recorded as -50dBm. During the second period, the mobile device may have recorded signal strength between -85 and -80, indicating that the mobile device did not enter store B. Therefore, the stop detection utility can specify a visit state of “in progress” or “stopped” for the store B. In response to the “in progress” visit state, the outage detection utility may not collect and store sensor data during the second time period. During the third time period, the outage detection utility retrieves accelerometer data, one or more electronic messages (e.g., text or email advertisements, coupons, event schedules, receipts, etc.), and GPS coordinates over the polling period. It can be used to identify that the mobile device was in proximity to store C. For example, the stop detection utility reports that the mobile device was traveling away from Store C at 3.5 mph at 12:05 pm, but the mobile device received an email ad for Store C at 12:06 pm, and the mobile device At 08pm, he changed his course and proceeded toward Store C, and the mobile device was progressing toward Store C at 3.5 mph between 12:08 and 12:15. C's Wi-Fi `` Store C '' detected and mobile device can be identified as traveling between 0.1 and 1.8 mph (e.g. wandering speed) between 12:15 pm and 12:45 pm It is. Based on this data, the stop detection utility can estimate the visit state of “visiting” for store C. In response to the visiting state of “visiting,” the outage detection utility may collect and store sensor data during the third time period.

  In operation 304, a set of candidate sites can be generated. In aspects, the mobile device sensor data may be accessible to a field retrieval utility, such as retrieval engine 202. The site retrieval utility can analyze the sensor data to identify the site and / or associated site data. Site data can be applied to models or algorithms that can be used for site identification. For example, the field data can be applied to a classification algorithm, such as a k-nearest neighbor algorithm. The classification algorithm may use the site data to identify candidate sites that are within a particular proximity or density distribution of the location reported for the mobile device. For example, a classification algorithm may utilize a geographic mapping service to identify all sites within 500 feet from a set of geographic coordinates. In at least one example, the classification algorithm may further include factors such as site popularity, how recent site visits were, site ratings, sales data (regional, seasonal, etc.), user preference data, etc. It is possible to incorporate. For example, a classification algorithm may use a knowledge source to retrieve popularity information about one or more identified venues. The sites can then be sorted according to site popularity, and the most relevant / popular sites (e.g., the 100 most popular sites) can be selected .

In operation 306, the candidate site data may be characterized to generate one or more feature sets. In aspects, a dataset of candidate sites may be provided to or accessible by a characterization component, such as feature engine 204. The characterization component can identify sensor data and / or site data corresponding to one or more of the candidate sites. The sensor / site data can be evaluated to generate a set of features indicative of the attributes (and corresponding values) associated with the candidate site. For example, the characterization component may receive as input the following sensor data for one or more sites.
[{'frequency': 2412,
'macaddress': u'e2: 55: 7d: 3f: 4b: e3',
'signalstrength': -63,
'ssid': u'IIDI ',
'timestamp': 1467215627},
{'frequency': 2412,
'macaddress': u'e2: 55: 7d: 3f: 4b: e2',
'signalstrength': -63,
'ssid': u'IDEAL-GUEST ',
'timestamp': 1467215627},
{'frequency': 2462,
'macaddress': u'54: 3d: 37: 3e: 03: 18',
'signalstrength': -78,
'ssid': u'Thrillist ',
'timestamp': 1467215627}]

The characterization component can generate a set of scores corresponding to one or more elements or features in the sensor data. For example, the characterization component may generate one or more location-based scores using the sensor data described above and / or historical data related to the scene represented by those sensor data. Is possible. In certain examples, a location-based score may represent an evaluation of various Wi-Fi networks. For example, for the sensor data described above, the following location-based scores can be generated.
[{Venue A:
wifi_match_score_1: 0.50,
wifi_match_score_2: 0.20},
{Venue B:
wifi_match_score_1: 0.90,
wifi_match_score_2: 0.95},
{Venue C:
wifi_match_score_1: 0.35,
wifi_match_score_2: 0.40}]

  The set of sensor data and scores can be used to generate and / or populate one or more feature vectors. For example, the characterization component can organize a set of features and values into one or more types or categories, such as location, device, time, movement, and so on. Organized features and feature values can be scaled and / or normalized. The normalized features and / or feature values can then be added to a feature vector. The feature vector can be provided to one or more data analysis utilities and / or data stores, or made accessible to one or more data analysis utilities and / or data stores.

  At operation 308, one or more metrics may be generated for the set of candidate sites. In aspects, the feature vectors may be accessible by a data analysis utility, such as the stochastic model engine 206. The feature vector may include feature data and / or a feature score for a candidate set of sites. The data analysis utility can generate a stochastic model using one or more gradient boosting techniques. The probabilistic model may evaluate the feature vectors to generate one or more probability or confidence scores for each site in the candidate set of sites. The probability / confidence score may indicate the likelihood that the scene corresponds to a particular set of coordinates, or the probability that the scene is being visited. In some examples, generating a probability / reliability score can include applying a scoring algorithm or machine learning technique to the feature scores in the feature vector corresponding to each candidate site. . In a particular example, the scoring algorithm can include summing the feature scores of each candidate site. In another example, a scoring algorithm may include applying a set or multiplier of weights to one or more feature scores. In aspects, the candidate sites may be organized or ranked, for example, according to a probability / reliability score generated by the probability model.

  In operation 310, the site list may be calibrated to increase the accuracy of the site list. In aspects, a calibration component, such as calibration engine 210, can access the site listing. The calibration component can apply one or more processing techniques to the site list to increase the accuracy and / or reliability metrics of the site selection or reduce the number of site selections. For example, the calibration component may compare one or more probabilities or confidence scores associated with the candidate site to a confidence threshold. The reliability threshold may represent a probability / reliability score that a site candidate must exceed in order to be included in the site list. In some examples, a threshold may be specified or selected, for example, by applying one or more sets of heuristics to the candidate site. Heuristics can be used to generate a site visit distribution for a candidate site. The site visit distribution can identify the frequency with which site visits were explicitly or implicitly reported during a period of time. In some aspects, a site visit distribution for a site candidate can be compared to a site visit distribution for training data. A threshold can be selected based on the evaluation. Candidate sites below the threshold can be removed from the list of candidate sites or significantly reduced. For example, a candidate site having a confidence score of less than 1% can be removed from the list of candidate sites, and a candidate site having a reliability score of less than 50% can be removed accordingly. It can be marked. Marking can include modifying the color, font, or transparency of the candidate site, or adding an indicator to the candidate site.

  In embodiments, after the site list has been calibrated (or as part of the calibration process), the calibration component may generate a list of the top "N" sites. The list of top "N" sites may include one or more candidate sites from the site list. The list of top "N" sites may indicate that those site candidates are most likely to correspond to a mobile device location or visit event. In some examples, the top "N" sites are selected by selecting a specific number of site results, selecting site results that exceed a certain probability threshold, evaluating user-selected data, and so on. It can be specified. In some embodiments, the list of top "N" sites can be ranked and / or organized according to one or more criteria. For example, a site listing might include "Store A" with a 55% preliminary reliability score, "Store B" with a 40% preliminary reliability score, and 5% It is possible to include “Store C” having a high reliability score. After this site list has been curated, the curated site list has “Store A” with a final reliability score of 25%, and “Store A” with a final reliability score of 75%. It is possible to include "Store B" and "Store C" having a final reliability score of 0%. The calibration component is capable of calibrating a curated site list, whereby sites are listed in the order of store A, store B, and store C. In at least one example, the curated site list can be ranked according to the final confidence score, whereby store B is the highest ranked site and store A Is a site ranked second. A setting that specifies that the top two sites will be listed in the curated site list can cause the site of store C to be removed from the curated site list. In some aspects, the curated venue list may be presented to the user via an interface, such as user interface 208.

  FIG. 4 illustrates an exemplary suitable operating environment for the field sensing system described in FIG. Operating environment 400 typically includes at least one processing unit 402 and memory 404 in its most basic configuration. Depending on the exact configuration and type of computing device, memory 404 (containing instructions for performing the field sensing embodiments disclosed herein) may be volatile (e.g., RAM), It can be non-volatile (ROM, flash memory, etc.), or some combination of the two. This most basic configuration is shown in FIG. In addition, environment 400 can include storage devices (removable 408 and / or non-removable 410), including but not limited to magnetic or optical disks or tape. Similarly, environment 400 may have input devices 414, such as a keyboard, mouse, pen, voice input, etc., and / or output devices 416, such as displays, speakers, printers, etc. One or more communication connections 412, such as a LAN, WAN, point-to-point, etc., may also be included in the environment. In embodiments, the connections can be operable to facilitate point-to-point communication, connection-oriented communication, connectionless communication, and the like.

  Operating environment 400 typically includes at least some form of computer readable media. Computer readable media can be any available media that can be accessed by processing unit 402 or other device that makes up the operating environment. By way of example, and not limitation, computer readable media may include computer storage media and communication media. Computer storage media includes volatile and non-volatile media, removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. And non-removable media. Computer storage media can be RAM, ROM, EEPROM, flash memory, or other memory technology, CD-ROM, digital versatile disk (DVD), or other optical storage, magnetic cassette, magnetic tape, magnetic disk storage, or other Or any other non-transitory media that can be used to store desired information. Computer storage media does not include communication media.

  Communication media embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transmission mechanism and includes any information delivery media. The term "modulated data signal" means a signal that has one or more of its characteristics set or changed in such a manner as to encode information in the signal. By way of example, and not limitation, communication media includes wired media such as a wired network or direct-wired connection, and wireless media such as acoustic, RF, infrared, microwave and other wireless media. Combinations of any of the above also are within the scope of computer readable media.

  Operating environment 400 can be a single computer operating in a networked environment using logical connections to one or more remote computers. The remote computer can be a personal computer, server, router, network PC, peer device, or other common network node, and typically includes the elements described above as well as so-called. Not including many or all of the other elements. Logical connections can include any of the methods supported by available communication media. Such networking environments are common in offices, enterprise-wide computer networks, intranets, and the Internet.

  The embodiments described herein may be employed using software, hardware, or a combination of software and hardware to implement and perform the systems and methods disclosed herein. It is possible to be. Although specific devices are listed throughout the disclosure as performing certain functions, these devices are provided for illustrative purposes and may be used without departing from the scope of the disclosure. One of ordinary skill in the art will appreciate that other devices can be employed to perform the functionality disclosed herein.

  The present disclosure describes some embodiments of the present technology with reference to the accompanying drawings, in which only some of the possible embodiments are shown. However, other aspects can be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of possible embodiments to those skilled in the art.

  Although particular embodiments are described herein, the scope of the technology is not limited to those particular embodiments. One skilled in the art will recognize other embodiments or modifications that fall within the scope and spirit of the technology of the present invention. Thus, the specific structure, acts, or media are disclosed only as exemplary embodiments. The scope of this technology is defined by the appended claims and any equivalents therein.

100 systems
102 Client device
104 Distributed Network
106 Field analysis system
108 Storage
200 input processing device
202 retrieval engine
204 Features Engine
206 Stochastic Model Engine
208 User Interface
210 Calibration Engine
300 methods
400 Operating environment
402 Processing Unit
404 memory
408 removable storage device
410 Non-removable storage device
412 Communication connection
414 Input Device
416 output device

Claims (20)

One or more processors,
A memory coupled to at least one of the one or more processors, the memory including computer-executable instructions, wherein the computer-executable instructions are executed by the at least one processor. Perform a method for on-site detection when performed, wherein the method comprises:
Receiving sensor data from a mobile device, wherein the sensor data is associated with a location of the mobile device;
Generating a site list using the sensor data, wherein the site list includes one or more candidate sites corresponding to the location of the mobile device;
Characterizing site data associated with the one or more candidate sites to generate a feature set;
Generating a metric for the one or more candidate sites by applying the feature set to a probability model;
Calibrating the site list based in part on the metric.   The system of claim 1, wherein the sensor data comprises at least one of geolocation data, location data, Wi-Fi information, software information, hardware information, accelerometer data, and time information.   The system of claim 1, wherein the sensor data indicates a visit by the mobile device to the one or more sites. Characterizing the site data,
Identifying one or more features in the field data;
Generating said feature set using said one or more features.   Identifying one or more features assesses at least one of the age of the site, site popularity, proximity to other sites, historical accuracy of site candidates, and previous site visit data 5. The system of claim 4, comprising the step of:   The system of claim 1, wherein the probabilistic model is a collection of decision trees generated using one or more gradient boosting techniques.   The probability model identifies a reliability score for one or more sites in the set of sites, wherein the reliability score indicates a probability that a corresponding site corresponds to the location of the mobile device. 6. The system according to 6.   The method of claim 1, wherein generating the site list includes identifying at least one site information within the sensor data and providing a set of geographic coordinates representing the location to a site identification utility. System. The method comprises
Accessing a set of training data including previously received input data and corresponding output data;
Providing output data generated by the probability model to an objective function;
Using the objective function to compare the output data generated by the stochastic model with the output data in the training data.   Calibrating the site list includes selecting the top `` N '' sites, wherein the top `` N '' sites are most likely to correspond to the location of a mobile device. The system of claim 1, wherein the system represents a site. The method comprises
Presenting the site list using a user interface;
Receiving, via the user interface, user input associated with the presented venue list;
Modifying the site list using the user input.   The system of claim 1, wherein the sensor data is received in response to detecting at least one of a stop event by the mobile device and a visit event by the mobile device. A method for on-site detection,
Receiving sensor data from a mobile device, wherein the sensor data is associated with a location of the mobile device;
Generating a site list using the sensor data, wherein the site list includes at least a first site and a second site, wherein the first site and the second site are the mobile sites. Steps associated with said location of the device;
Characterizing site data associated with the first site and the second site to generate a feature set;
Generating a reliability metric for the first site and the second site by applying the feature set to a probability model;
Calibrating the site list based in part on the metric;
Presenting, via a user interface, the calibrated site list and at least a portion of the reliability metric, wherein at least one of the first site and the second site includes: Corresponding to the location of a mobile device.   14. The method of claim 13, wherein the probabilistic model is a set of decision trees generated using one or more gradient boosting techniques.   The one or more gradient boosting techniques may include at least one of site age, site popularity, proximity to other sites, historical accuracy of site candidates, previous site visit data, and user feedback. 15. The method of claim 14, wherein one is evaluated.   14. The method of claim 13, wherein the sensor data indicates a visit by the mobile device to at least one of the first site and the second site. Accessing a set of training data including previously received input data and corresponding output data;
Providing output data generated by the probability model to an objective function;
Using the objective function to compare the output data generated by the probability model with the output data in the training data;
Scoring the validity of the probability model based on the comparison.   17. The method of claim 16, wherein the reliability metric indicates a probability that each site corresponds to the location of the mobile device.   Generating the site list includes identifying a first set of sites that are within a first proximity to the location of the mobile device, wherein calibrating the site list comprises: Identifying a second set of sites that are within a second proximity to the location of the mobile device, wherein the second set of sites includes less sites than the first set of sites. 14. The method of claim 13, comprising. A computer-readable storage medium encoding computer-executable instructions, wherein the computer-executable instructions, when executed by at least one processor, perform a method for field detection, the method comprising:
Receiving sensor data from a mobile device, wherein the sensor data is associated with a location of the mobile device;
Generating a site list using the sensor data, wherein the site list includes at least a first site and a second site, wherein the first site and the second site are the mobile sites. Steps associated with said location of the device;
Characterizing site data associated with the first site and the second site to generate a feature set;
Generating a reliability metric for the first site and the second site by applying the feature set to a probability model;
Calibrating the site list based in part on the metric;
Presenting, via a user interface, the calibrated site list and at least a portion of the reliability metric, wherein at least one of the first site and the second site includes: Corresponding to the location of a mobile device.

Images